Integration of Transformer Models for Time-Series Forecasting

Abstract

This paper investigates the use of transformer models for time-series forecasting in the energy market, particularly focusing on inter-provincial spot prices. By leveraging the transformer’s self-attention mechanism, the proposed model captures long-range dependencies more effectively than traditional models such as ARIMA and LSTM. Our research highlights the significance of improved forecasting accuracy for stakeholders, including utilities, policymakers, and energy traders. The proposed approach is evaluated under diverse market conditions, including stable and volatile periods, to emphasize its robustness. Results demonstrate significant performance improvements, particularly in volatile markets, showcasing the model's potential for optimizing purchasing strategies and enhancing energy market stability